Most enterprises are not failing at AI because of the technology. They are failing because they are applying powerful tools to flawed processes and expecting different results. The organisations pulling ahead have understood a different imperative: redesign the work first, then embed AI into it.

The Adoption Paradox

Enterprise AI adoption has never been higher. According to McKinsey's 2025 State of AI survey, 88 per cent of organisations now use AI in at least one business function, up from 78 per cent the year before. Boards are engaged. Budgets have been approved. Proof of concepts have been run.

And yet, only around six per cent of those organisations are seeing meaningful financial impact at an enterprise level.

That gap does not exist because the models are underperforming. It exists because most organisations are pointing AI at the wrong target. They are automating the work as it exists, rather than reimagining how the work should be done.

42% of companies abandoned most of their AI initiatives in 2025, up from 17% the previous year. (S&P Global Market Intelligence, 2025)

Automating the Problem, Not Solving It

There is a well-known principle in process improvement: if you automate a broken process, you get a faster broken process. AI amplifies this effect significantly.

Many enterprise AI initiatives are structured around existing workflows. The process stays the same; AI is layered on top to handle specific steps or reduce manual effort at particular points. The result tends to be marginal efficiency gains, mounting complexity, and a growing sense that the investment is not delivering on its promise.

This happens because the way a process was designed often reflects the constraints of how work was done before AI existed. Approval chains were built around manual checks. Sequential handoffs were necessary because systems could not share data in real time. Steps were repeated because there was no single source of truth. Applying AI on top of these structures does not remove the underlying constraints; it just makes the journey through them slightly quicker.

The question worth asking is not, "How can AI speed up step three?" It is, "Does step three need to exist at all?"

What the Research Tells Us

The evidence on this point is consistent across multiple independent research programmes.

McKinsey tested 25 organisational attributes across nearly 2,000 companies and found that fundamental workflow redesign had the single strongest correlation with EBIT impact from AI. High-performing organisations were almost three times more likely to have redesigned their workflows end to end, rather than layering AI onto existing processes.

BCG research reaches the same conclusion. Companies that deploy AI tools without changing how work gets done see minimal returns. Those that reshape workflows end to end shift employees toward higher-value tasks and save significantly more time in the process.

RAND Corporation found that AI projects fail at roughly twice the rate of other IT projects. The most common reasons were not technical. Projects failed because teams optimised the wrong metric, the AI did not fit the actual workflow, or the initiative was driven by enthusiasm for the technology rather than clarity on the business outcome.

MIT research found that 95 per cent of enterprise AI pilots delivered zero measurable financial return. Again, the underlying technology was rarely the issue. The failure points were workflow integration, data readiness, and the absence of a clearly defined outcome before the build began.

3x AI high performers are nearly three times more likely to have fundamentally redesigned their workflows, compared to their peers. This is the single strongest predictor of meaningful AI value. (McKinsey State of AI, 2025)

What Workflow Redesign Actually Means

Workflow redesign is not a technology project. It is a business transformation project that uses technology as the enabler.

In practical terms, it means taking a function, such as finance, operations, procurement, or HR, and asking a fundamentally different question: if we were designing this from scratch today, knowing what AI can do, how would the work flow? Which decisions could be made automatically? Which handoffs could run in parallel rather than in sequence? Which steps only existed because of legacy system constraints that no longer apply?

This is a materially different exercise from process mapping followed by automation. It requires people who understand both the function and the technology, not just one or the other. A pure technologist will optimise what is there. A pure domain expert may not know what AI makes possible. The value is in the combination.

Sequential to Parallel: A Simple but Powerful Shift

One of the clearest practical changes that comes from reimagining workflows is the shift from sequential to parallel working. Traditional enterprise processes tend to be linear by design: step one must be completed before step two can begin, which must finish before step three is triggered.

With AI, many of those dependencies disappear. Data can be gathered and validated simultaneously across multiple sources. Risk checks can run alongside approval processes rather than after them. Downstream activities can begin as early signals emerge, rather than waiting for the entire upstream process to complete.

The operational benefit of this shift is substantial. Time that was previously spent waiting, chasing, and re-checking becomes time spent on decisions and outcomes. But this kind of redesign cannot happen without challenging the assumptions baked into the original process.

Some Processes Should Not Be Automated at All

Genuine workflow redesign will sometimes conclude that a process, or part of one, should simply be removed rather than automated. Steps that existed purely as manual workarounds for system limitations no longer need to exist. Approval stages that were designed for a world without real-time data may be redundant. Reporting layers built for manual aggregation become unnecessary when data flows automatically.

This is, understandably, a more uncomfortable conversation than automation. It requires people to let go of familiar ways of working and, sometimes, to acknowledge that the current process was built around constraints rather than around the best outcome. That conversation requires trust between the organisation and its partners, and it requires partners who are willing to have it.

Why Functional Expertise Cannot Be Optional

A significant part of why AI projects underdeliver is that the teams running them understand the technology but not the function. They can implement a model, build a pipeline, and connect APIs. What they often cannot do is look at an end-to-end operations or finance process and identify where the real drag is, what the downstream consequences of a change would be, or how the people doing the work will respond to a fundamentally different way of working.

Accenture's 2025 enterprise report noted that 97 per cent of executives believe generative AI will transform their company, but 65 per cent say they lack the expertise to lead those transformations. The gap is not in appetite; it is in the capacity to execute functional change.

The organisations that are succeeding with AI at scale are combining domain knowledge in operations, finance, commercial, and HR with a rigorous understanding of what AI can actually do in those contexts. They are not asking technologists to lead functional change programmes, and they are not asking functional teams to make technology decisions in isolation.

The most valuable external partners in this space are those who can sit alongside both sets of people and bridge the gap between them.

Why So Many Organisations Are Stuck in Pilot Mode

One of the defining characteristics of the current AI landscape is what is sometimes called pilot purgatory: organisations that have run multiple proofs of concept, can point to individual examples of AI working well, but cannot demonstrate enterprise-level return on their investment.

This is almost always a consequence of the automation-first approach. Individual steps get improved. Isolated tasks get faster. But the overall process remains the same, and the cumulative value of a dozen small improvements rarely adds up to a meaningful business outcome.

Gartner has predicted that 60 per cent of AI projects lacking proper workflow integration will be abandoned before they reach scale. Given that S&P Global recorded a 42 per cent abandonment rate in 2025 alone, that trajectory appears to be tracking.

Getting out of pilot purgatory requires a deliberate decision to pursue functional transformation rather than incremental automation. It requires identifying a business function, scoping the end-to-end process, and committing to redesigning how that work is done with AI at its centre, not bolted on at the edges.

What the Right Approach Looks Like in Practice

The organisations achieving meaningful returns from AI share a consistent pattern of behaviour. They are not necessarily the ones with the largest AI budgets or the most technically sophisticated teams.

1. They start with a clearly defined business outcome, not a technology. The question is always what needs to change in terms of speed, cost, accuracy, or customer experience, not which AI model to deploy.

2. They treat AI implementation as a cross-functional transformation programme, not an IT project. This means involving the people who actually do the work, not just the people who manage the systems.

3. They challenge existing processes before they design the AI layer. Steps that should be eliminated are eliminated. Handoffs that should be parallelised are parallelised. Only then is AI applied to what remains.

4. They measure outcomes at a business level, not just at a task level. Cost reduction per transaction, cycle time from request to resolution, error rate on a specific process: these are the metrics that matter, not the number of AI tools deployed.

5. They work with partners who can speak to both the functional and the technical dimensions of the transformation, and who can demonstrate concrete examples of what they have actually delivered, not just what they are capable of in principle.

The Shift That Changes Everything

The organisations that will look back on 2025 and 2026 as the moment they pulled ahead of their competitors will not be the ones that ran the most AI pilots. They will be the ones that made a deliberate decision to stop automating the present and start redesigning the future.

That shift requires asking harder questions. It requires bringing functional expertise into the room alongside technical capability. It requires the willingness to change or eliminate processes that have existed for years, and to measure success in terms of business outcomes rather than technology outputs.

It is not the easiest path. But it is the only one that reliably leads to AI transformation rather than AI adoption.

About VE3

VE3 is a UK-based enterprise AI, data, and digital transformation consultancy and Microsoft Solutions Partner. We work with organisations across the public and private sector to design and deliver AI-enabled transformation that creates measurable business value. Our approach combines deep functional domain expertise with technical delivery capability, so that the work we do changes how organisations operate, not just the tools they use